← Back to Attention Labs
BBB-Nuke × Claude

Use BBB-Nuke with Claude.

BBB-Nuke ships as a Model Context Protocol server, which means Claude can call it as a native tool inside any conversation. Paste a SMILES, upload a CSV, or describe an analog series, and Claude reasons over the scores, flags efflux liabilities, and drafts the next round — all without leaving the chat.

What you'll need.

Claude Code (one-liner).

Run this in your terminal and you're done. Claude Code picks up the server on its next session.

claude mcp add bbnuke --transport http https://mcp.attentionlab.ai/mcp

Claude Desktop (config file).

Open Claude Desktop → Settings → Developer → Edit Config, and add the bbb-nuke entry under mcpServers:

{
  "mcpServers": {
    "bbnuke": {
      "type": "url",
      "url": "https://mcp.attentionlab.ai/mcp"
    }
  }
}

The config file lives at ~/Library/Application Support/Claude/claude_desktop_config.json on macOS and %APPDATA%\Claude\claude_desktop_config.json on Windows. Restart Claude Desktop after saving.

Verify the server is live.

Start a new Claude chat and ask:

“What MCP tools do you have available from bbb-nuke?”

Claude should list the BBB-Nuke scoring tools. If nothing shows up, jump to Troubleshooting.

Try these prompts.

Score the molecule CN1CCC[C@H]1c1cccnc1 for blood-brain barrier penetration.
Here are 20 SMILES from my ADHD analog series. Screen them with BBB-Nuke and flag anything that looks like an efflux liability.
Compare CNS-MPO vs BBB-Nuke scores for this compound and explain the difference.
Draft three analogs of this lead that improve the BBB-Nuke score without losing the core scaffold.

Troubleshooting.

Claude doesn't see the bbb-nuke tools.

Fully quit and relaunch Claude Desktop. On Claude Code, run claude mcp list to confirm the server is registered.

401 / invalid API key.

Double-check that BBB_NUKE_API_KEY is set in the server's env block (not in your shell). Keys are issued per-account and are not interchangeable between dev and prod.

Rate limited.

Default accounts are capped at 1,000 compounds per 24 hours. If you need higher throughput or a dedicated quota, contact us about Enterprise.

Something else.

Email temi@attentionlab.ai with the error and your client (Desktop or Code).

Need API access, higher throughput, or an on-prem deployment?

temi@attentionlab.ai